dsh-better-input

If you talk with agents often, you may still feel that input is awkward: typing everything by hand, pasting rough spoken notes, and then cleaning them up yourself. dsh-better-input is built to reduce that friction. It turns the DeepSeek Harness input area into a smoother entry point where speech becomes text, drafts can be polished automatically, prompts can be optimized, and local files can enter the conversation more naturally.

Its strength is not a single feature, but a whole input experience centered on the content you feed to an agent. Browser-based speech recognition streams your words into the input box. AI polishing removes filler words, fixes homophone errors from speech recognition, adds punctuation, and converts spoken lists into structured text. Prompt optimization helps you express intent more precisely before sending. When you have local documents, spreadsheets, presentations, web pages, or ebooks, it can convert them into clear Markdown so the AI can understand the context more easily; for scanned pages without a text layer, it also offers an OCR path through a configured vision model. These capabilities reuse DSH model configurations as much as possible, lowering extra setup cost.

It is a good fit for people who want a lighter input experience: users who capture notes by voice, people who paste code or text fragments into chats, users who need to bring local documents into agent conversations, and anyone who wants better prompts without spending too much attention on formatting and cleanup.